Update app.py

app.py

@@ -48,7 +48,20 @@ def transliterate_to_sinhala(text):
 # conv_model_name = "microsoft/Phi-3-mini-4k-instruct"  # Use GPT-2 instead of the gated model
 # tokenizer = AutoTokenizer.from_pretrained(conv_model_name, trust_remote_code=True)
 # model = AutoModelForCausalLM.from_pretrained(conv_model_name, trust_remote_code=True).to(device)
-pipe1 = pipeline("text-generation", model="TinyLlama/TinyLlama-1.1B-Chat-v1.0").to(device)
+# pipe1 = pipeline("text-generation", model="TinyLlama/TinyLlama-1.1B-Chat-v1.0").to(device)
+
+model = "tiiuae/falcon-7b-instruct"
+
+tokenizer = AutoTokenizer.from_pretrained(model)
+text_gen_pipeline = transformers.pipeline(
+    "text-generation",
+    model=model,
+    tokenizer=tokenizer,
+    torch_dtype=torch.bfloat16,
+    trust_remote_code=True,
+    device_map="auto",
+)
+
 
 # client = InferenceClient("google/gemma-2b-it")
 
@@ -72,8 +85,19 @@ def conversation_predict(text):
     # outputs = model.generate(**input_ids)
     # return tokenizer.decode(outputs[0])
 
-    outputs = pipe1(text, max_new_tokens=256, temperature=0.7, top_k=50, top_p=0.95)
-    return outputs[0]["generated_text"]
+    # outputs = pipe1(text, max_new_tokens=256, temperature=0.7, top_k=50, top_p=0.95)
+    # return outputs[0]["generated_text"]
+
+    sequences = text_gen_pipeline(
+        text,
+        max_length=200,
+        do_sample=True,
+        top_k=10,
+        num_return_sequences=1,
+        eos_token_id=tokenizer.eos_token_id,
+    )
+    return sequences[0]['generated_text']
+
 
 def ai_predicted(user_input):
     if user_input.lower() == 'exit':